Portable MCP Toolkit

# Verifying the MCP Fix Works ## Quick Verification Steps ### Step 1: Verify Configuration Check that `.cursor/mcp.json` uses the `cmd /c` wrapper: ```json { "mcpServers": { "code-intelligence": { "command": "cmd", "args": [ "/c", "python", "C:\\AIStack-MCP\\mcp_intelligence_server.py", "--workspace", "${workspaceFolder}" ], "env": { "OLLAMA_URL": "http://localhost:11434", "QDRANT_URL": "http://localhost:6333" } } } } ``` **Important:** Update `C:\\AIStack-MCP\\` to match your actual installation path. ### Step 2: Run Automated Test Run the PowerShell test script: ```powershell .\test_mcp_fix.ps1 ``` This will verify: - ✅ Configuration uses `cmd /c` wrapper - ✅ Python and dependencies are installed - ✅ Ollama and Qdrant services are running - ✅ Server can start successfully - ✅ `cmd /c` wrapper works correctly ### Step 3: Manual Test (If Automated Test Passes) Test the server directly with `cmd /c`: ```powershell # This should start the server and wait for input (expected behavior) cmd /c python C:\AIStack-MCP\mcp_intelligence_server.py --workspace C:\AIStack-MCP ``` **Expected behavior:** - Server prints initialization logs - Shows "Ready for MCP connections" - **Hangs waiting for JSON-RPC input** (this is CORRECT - means STDIO is working) - Press `Ctrl+C` to stop **If you see errors:** Fix them before testing in Cursor. ### Step 4: Test in Cursor 1. **Close Cursor completely** (all windows) 2. **Reopen Cursor** 3. **Open this workspace** (`C:\AIStack-MCP`) 4. **Wait 5-10 seconds** for MCP server to initialize 5. **Check if MCP tools are available:** - Open AI chat - Look for MCP tools in the tool list - Tools should include: `semantic_search`, `analyze_patterns`, `get_context`, `generate_code`, `index_workspace` ### Step 5: Verify Tools Work Try using a tool: ``` @semantic_search query: "How does error handling work?" ``` **If tools appear and work:** ✅ Fix is successful! **If Cursor still crashes or hangs:** 1. **Check Cursor logs:** - Press `Ctrl+Shift+P` (or `Cmd+Shift+P` on Mac) - Type "Toggle Developer Tools" - Go to Console tab - Filter for "code-intelligence" or "mcp" - Look for error messages 2. **Common issues:** - `ECONNREFUSED` → Ollama or Qdrant not running - `ENOENT` → Python path incorrect in mcp.json - `ETIMEDOUT` → STDIO pipe still not working (try alternative solutions) - `ClosedResourceError` → Windows STDIO issue (try virtualenv solution) 3. **Try alternative solutions:** - Use virtualenv config: Copy `.cursor/mcp.json.example.virtualenv` to `.cursor/mcp.json` - Use uv config: Copy `.cursor/mcp.json.example.uv` to `.cursor/mcp.json` ## Success Indicators ✅ **Server starts without errors** (logs show successful initialization) ✅ **Cursor doesn't crash or hang** when opening workspace ✅ **MCP tools appear** in AI chat interface ✅ **Tools can be called** and return results ✅ **No errors in Cursor console** related to MCP ## Failure Indicators ❌ **Cursor crashes immediately** when opening workspace ❌ **Cursor hangs/freezes** for more than 30 seconds ❌ **MCP tools don't appear** in AI chat ❌ **Error messages in Cursor console** about MCP connection ❌ **Server logs show errors** before "Ready for MCP connections" ## Next Steps After Verification If the fix works: 1. ✅ Document your working configuration 2. ✅ Note any path customizations you made 3. ✅ Consider committing the working `mcp.json` (if not in .gitignore) If the fix doesn't work: 1. 🔍 Review Cursor console logs for specific errors 2. 🔄 Try alternative solutions (virtualenv or uv) 3. 📝 Document the specific error messages 4. 🐛 Report issue with full error details